The 2018 flood experienced by Kerala, the southernmost state of India, has affected thousands of lives, resulting in heavy financial loss (248,930 million INR) and structural damage (129,420 residential buildings have been reported to be damaged). Depth versus damage relationships have been used conventionally for flood emergency decision-making; however, these are approximate and do not account for all the significant factors involved in the flood damage assessment. An extensive questionnaire survey was conducted in this study to collect data comprising significant parameters of flood-damaged residential buildings due to the 2018 Kerala flood event. The damage to the buildings was expressed in terms of damage ratios, and the buildings were classified into various damage states according to their damage ratios. Machine learning (ML) techniques, such as random forest, naïve Bayes, decision tree, K-nearest neighbors, AdaBoost, XGBoost, support vector machine, LightGBM, and CatBoost, were used in this study to develop models to predict and classify the buildings into various flood damage states. Results indicated that almost all the tree-based ML techniques performed well, and the random forest model was obtained as the best-performing classification model for flood damage prediction with 84% accuracy. Further SHAP (SHapely Additive exPlanations) analysis was conducted to explore the relative importance of the different parameters involved in damage prediction.